BofA sees server CPU TAM hitting $170bn by 2030 as NVIDIA takes on AMD

Investing.com -- Bank of America Research sees the server CPU total addressable market reaching $170 billion by 2030, roughly four times current levels, as NVIDIA's debut of its Vera CPU architecture ignites a direct contest with AMD over how agentic AI workloads should be measured and monetized.

BofA maintains a Buy rating and $350 price target on NVIDIA (NASDAQ: NVDA), framing the Vera launch as the opening salvo in what it considers the defining infrastructure debate of the current AI cycle. Advanced Micro Devices (NASDAQ: AMD) is the direct competitive counterpart, with its EPYC server line and Thursday's AI 2026 Day serving as the immediate market test of which performance philosophy wins enterprise adoption.

The central question BofA poses is one of architectural philosophy, not raw specs. "The key question for investors is whether agentic AI is primarily constrained by time-to-complete an agent or number-of-agents-per-rack," analyst Vivek Arya wrote. NVIDIA's Vera CPU is built around the former view: it combines 88 custom Olympus ARM-based cores, 1.2TB/s memory bandwidth, and 3.4TB/s on-die fabric bandwidth, and it is positioned as part of a co-designed system spanning six AI building blocks, including the Rubin GPU, Groq LPX, Spectrum switches, and BlueField storage and network interface cards. The monolithic compute die at Vera's core, which NVIDIA claims provides scalable coherency, stands in deliberate contrast to AMD's proven chiplet approach.

AMD's numbers on a raw throughput basis are harder to dismiss. BofA notes that AMD estimates its EPYC 9965 (Turin) chip delivers approximately 2.4x the rack-level throughput of NVIDIA's Vera baseline in a modeled 100kW deployment, with the next-generation EPYC 6 (Venice) projected to extend that advantage to roughly 3.3x total throughput, according to BofA's characterization of AMD's own estimates. Those figures anchor AMD's competing framework. "NVDA's framework is rooted in latency, per-thread progress, and GPU utilization. AMD's framework is rooted in concurrency, throughput, and service density," BofA wrote.

The firm is careful not to declare a winner. Instead, it sees AMD's AI 2026 Day on Thursday as a moment that may matter more for narrative than for benchmarks. "Our expectation is that AMD's Thursday event will be less about benchmark comparisons and more about establishing the industry's preferred metric," Arya wrote. Whichever framing — latency-per-agent or agents-per-rack — gains traction with hyperscalers and enterprise buyers will effectively set the performance standard against which future CPU investment is justified.

A secondary battleground is software incumbency. AMD and Intel both carry the weight of x86's installed base in enterprise environments, a structural advantage that BofA flags as potentially decisive as AI moves beyond model inference. "As AI expands from model inference into enterprise workflows, software incumbency could become an important differentiator," the firm noted, pointing to the ARM vs. x86 instruction set debate as a longer-term risk for NVIDIA's Vera, which runs on an ARM-based architecture.

For NVIDIA shareholders, the bull case embedded in BofA's $350 target rests on the co-design argument: Vera is not competing as a standalone CPU but as the connective tissue of an integrated AI system, with the Rubin GPU and the broader NVLink fabric amplifying its value in ways that raw throughput comparisons on a per-rack basis may not fully capture.

The immediate catalyst is Thursday's AMD AI 2026 Day, where the firm expects AMD to stake out its preferred language for measuring AI infrastructure performance. How hyperscalers and enterprise buyers respond to that framing, and whether they migrate toward AMD's density argument or NVIDIA's latency argument, will shape capital allocation in a server CPU market BofA projects will be four times larger within four years.

You May Also Be Interested In

• Wolfe upgrades AT&T to Outperform as improving fundamentals outweigh Starlink risk
• Critical Metals Corp reaches final round of Kenya rare earth tender
• System1 launches Dogpile Fetch, a search API for AI agents